Because vehicle number information is a key for identifying an individual vehicle, the vehicle number information is applied for various purposes. Here, a vehicle number is a character string assigned to the individual vehicle.
Because vehicles are targeted in vehicle number recognition in various environments such as a general road, a highway, a parking lot entrance, and the like, it is necessary to capture an unblurred precise image quickly without being affected by weather conditions or the like. Further, in the vehicle number recognition, an image recognition device for implementing an algorithm for accurately recognizing a vehicle number at a high speed and inexpensively using an obtained image as an input is essential.
As technologies related to this background, various technologies are known (for example, see Patent Literatures 1 and 2).
For example, in Patent Literature 1, a device for automatically identifying a vehicle number of a vehicle that is driving or stopped on a road is disclosed. More specifically, this device photographs a road. Then, the device determines a candidate for a vehicle number plate frame from a captured image. Then, the device determines a candidate for a vehicle number region from the image. Then, the device determines a vehicle number plate position based on an overlap degree between a part surrounded by the vehicle number plate frame estimated based on the candidate for the vehicle number plate frame and the vehicle number region estimated based on the candidate for the vehicle number region. In this manner, because this device can determine the vehicle number plate position from the overlap degree between the candidate for the vehicle number plate frame and the candidate for the vehicle number region, it is possible to perform a vehicle number plate segmentation process with a more reliable probability than when a determination of the vehicle number plate position is made from only a single candidate.
In addition, for example, in Patent Literature 2, a device for performing vehicle number plate recognition of a vehicle is disclosed. More specifically, this device reduces a fetched image in a horizontal direction and a vertical direction. Then, this device subtracts a value of a pixel of a position shifted by a predetermined amount in a positive or negative direction from each pixel from a value of each pixel in the reduced image, binarizes a subtraction result by comparing the subtraction result to a binary threshold value, and makes a pseudo shift and generates a correlation image in a logical product operation between a value at the time of the shift in the positive direction and a value at the time of the shift in the negative direction. Then, this device subtracts a value of a pixel of a position shifted by a predetermined amount in the positive or negative direction from each pixel from a value of each pixel in the reduced image, binarizes a subtraction result by comparing the subtraction result to a binary threshold value, and generates a pseudo logical sum image in a logical sum operation between a value at the time of the shift in the positive direction and a value at the time of the shift in the negative direction. Then, this device divides the pseudo-shift correlation image into a plurality of small regions, calculates an area of pixels of which luminance values are set to 1 among the small regions, and selects a small region of a vehicle number plate candidate based on an area value of each small region. Then, this device temporarily sets the vehicle number plate region on the pseudo logical sum image based on the small region of the vehicle number plate candidate and checks the suitability of the temporarily set region as the vehicle number plate region to segment the vehicle number plate region. In this manner, this device can segment the vehicle number plate region at a high speed and with high precision using a general-purpose calculation device without newly developing a dedicated calculation device and promote the reduction of a development period and the reduction of cost.